Food processing: quality by lab sample, freezers flat out

The same product on another line is tuned again from scratch

What happens. A product runs well on one line or in one plant. It moves to a second line with a different oven, filler or cooker, and the settings that worked do not fit. The start-up team adjusts temperatures, speeds and times again through trial batches.

What it costs. Scrap and rework during every transfer, line time spent on trials instead of saleable product, and experienced people tied up for days at another site.

Why the usual fix fails. Copying the settings across ignores that the lines differ in size, heat transfer and dead times. Correction tables built by hand cover the products that were moved before and break at the next new one.

How a fixed-price remote diagnosis checks it. From the recipes and process logs of both lines (temperatures, speeds, times, and the quality results that went with them) I check how far the difference between the lines can be described by a small model of each. The fixed-price remote diagnosis answers whether settings can be transferred through that model instead of being found again, and what data it would need.

Background: Replacing hand-tuned parameter tables by a model

Near-infrared calibrations break when the raw material changes

What happens. The near-infrared analyser on the line reads moisture, fat or protein well for months. Then a new supplier, a new harvest or a new recipe arrives, and the values drift away from the lab.

What it costs. Lab work and sample costs for every recalibration, and weeks in which the line is controlled on values nobody fully trusts. Wrong readings mean giveaway or off-spec product.

Why the usual fix fails. Adding the new samples to the calibration makes it fit again, until the next change. A purely statistical model only knows the raw materials it has seen; it has no idea what the process does to the product.

How a fixed-price remote diagnosis checks it. I look at your spectra, the lab values and the process data from the same periods, and check where the calibration fails and whether process knowledge (temperatures, flows, mixing ratios) explains the drift. The fixed-price remote diagnosis tells you whether combining spectra with a process model would hold across raw-material changes, and how many lab samples that would still need.

Background: Computed measurements

Chocolate temper is judged by a sample

What happens. The tempering machine runs on set temperatures, and the operator checks the temper with a sample, a temper meter or a smear on a knife. Between checks the cocoa butter crystals in the mass can change with the incoming chocolate, the room and the throughput.

What it costs. Under-tempered or over-tempered chocolate shows up later as bloom, poor gloss, sticking moulds or broken pieces. By then whole batches are moulded and packed, and returns or rework follow.

Why the usual fix fails. More frequent sampling helps but still looks backwards and takes staff time. Running cooler or longer for safety changes the viscosity and the coating weight, which brings its own problems.

How a fixed-price remote diagnosis checks it. From the temperatures, flows and motor loads of your tempering machines, the temper readings and the quality complaints, I check whether the temper state can be computed continuously from signals you already record. The fixed-price remote diagnosis shows how well such a computed value would have flagged the bad batches, and what it would take to use it on the line.

Freezing tunnels run at maximum whatever the load

What happens. The freezing tunnel or spiral freezer runs at the lowest air temperature and full fan speed, because the product must leave frozen to the core. Nobody measures the core temperature in line. Full or half empty, thick products or thin, the settings stay the same.

What it costs. Freezing is one of the largest power consumers in a frozen-food plant. Product that is already frozen gets more cold than it needs, and the refrigeration plant runs at full load in the most expensive hours.

Why the usual fix fails. Lowering the settings by hand is risky: one under-frozen batch means quality problems and possibly scrapped product. So the margin comes back.

How a fixed-price remote diagnosis checks it. From the tunnel data (air temperatures, fan speeds, belt speed, product type and load, refrigeration power) and your core-temperature checks I estimate how the core temperature develops through the tunnel. The fixed-price remote diagnosis tells you how much margin each product really has and whether a computed core temperature can lead the tunnel.

Background: Economic predictive control

Product changes are planned by hand

What happens. The planner builds the weekly sequence from experience: light before dark, allergen-free before allergen, small orders squeezed in where they fit. When an order changes or a line stops, the sequence is rearranged by phone and spreadsheet.

What it costs. Every change means cleaning, flushing, start-up waste and lost line time. A poor sequence adds changes that were not necessary, and urgent orders push the plan into extra shifts or late deliveries.

Why the usual fix fails. The planning software often holds orders and capacities but not the real cost of each change between two products. So the planner’s head stays the real tool, and the plan cannot be better than one person’s overview on a busy day.

How a fixed-price remote diagnosis checks it. From your production records (sequences, change and cleaning times, waste per change, order dates) I measure what the changes really cost and compare past plans with an optimised sequence for the same orders. The fixed-price remote diagnosis shows whether sequence optimisation would pay off and what it needs to fit your planning routine.

Background: Optimised schedules and setpoint plans

Key quality values have no inline measurement

What happens. Moisture, viscosity, fat content, degree of cooking or browning decide whether the product is right, but the line has no sensor for them. The lab measures samples a few times per shift. Process control works on temperatures and flows and hopes the product follows.

What it costs. Lab results arrive when the batch is already made. Off-spec product is found late, operators keep margins in the recipes, and the lab is busy with routine samples.

Why the usual fix fails. Physical sensors for these values are expensive, hard to clean, or do not survive the process. Projects to develop new sensors take years and often end with a prototype that never reaches the line.

How a fixed-price remote diagnosis checks it. I take the lab values you already have and the process data recorded at the same time, and check whether the quality value can be computed from them reliably. The fixed-price remote diagnosis shows how close such a computed measurement comes to the lab, where it fails, and whether it is good enough for control or only for alarms.

Background: Computed measurements

Low-temperature process heat still comes from fossil boilers

What happens. Cooking, pasteurising, cleaning and drying need hot water or steam at moderate temperatures, yet it comes from a gas or oil boiler sized for the peak. Waste heat from refrigeration and compressors goes out through the roof. Old boilers are often replaced like for like.

What it costs. Fuel and CO₂ costs that rise with every price increase, and heat the plant already has but throws away. A heat pump bought without knowing the real heat demand may run badly or sit idle.

Why the usual fix fails. Heat pumps and heat recovery are often planned from nameplate values and monthly bills. Real demand comes in peaks, at different temperatures and times; without that picture the system ends up oversized or short.

How a fixed-price remote diagnosis checks it. From your meter data, steam and hot-water flows, temperatures and production plans I build the heat picture of the site over a typical week: which demands, which waste heat, at which temperatures and when. The fixed-price remote diagnosis shows where heat integration and a heat pump fit, and how their control has to run against the production plan.

Background: Economic predictive control